Tensor Processing Primitives: A Programming Abstraction for Efficiency and Portability in Deep Learning & HPC Workloads

TL;DR

This paper introduces Tensor Processing Primitives (TPP), a versatile programming abstraction that enhances efficiency and portability in deep learning and high-performance computing workloads by using a compact set of tensor operators.

Contribution

The work presents TPP as a platform-agnostic yet highly optimized set of tensor operators, enabling portable and efficient implementation of complex DL workloads.

Findings

TPP-based kernels outperform state-of-the-art implementations.

End-to-end DL and HPC workloads using TPP achieve higher performance.

TPP provides a platform-agnostic programming model for tensor computations.

Abstract

During the past decade, novel Deep Learning (DL) algorithms, workloads and hardware have been developed to tackle a wide range of problems. Despite the advances in workload and hardware ecosystems, the programming methodology of DL systems is stagnant. DL workloads leverage either highly-optimized, yet platform-specific and inflexible kernels from DL libraries, or in the case of novel operators, reference implementations are built via DL framework primitives with underwhelming performance. This work introduces the Tensor Processing Primitives (TPP), a programming abstraction striving for efficient, portable implementation of DL workloads with high-productivity. TPPs define a compact, yet versatile set of 2D-tensor operators (or a virtual Tensor ISA), which subsequently can be utilized as building-blocks to construct complex operators on high-dimensional tensors. The TPP specification is platform-agnostic, thus code expressed via TPPs is portable, whereas the TPP implementation is highly-optimized and platform-specific. We demonstrate the efficacy and viability of our approach using standalone kernels and end-to-end DL & HPC workloads expressed entirely via TPPs that outperform state-of-the-art implementations on multiple platforms.